90 likes | 899 Views
Contents. Rao-Blackwell theoremMarginalizing the filter (Rao-Blackwellization)Generic Rao-Blackwellized Particle FilterPartially analytical state updatePractical RBPF AlgorithmMathematical Derivation. Rao-Blackwell theorem. Let d(X) be an estimator of an unobservable. X is the observable data.
E N D
1. Rao-BlackwellisedParticle Filter
2. Contents Rao-Blackwell theorem
Marginalizing the filter (Rao-Blackwellization)
Generic Rao-Blackwellized Particle Filter
Partially analytical state update
Practical RBPF Algorithm
Mathematical Derivation
3. Rao-Blackwell theorem Let d(X) be an estimator of an unobservable. X is the observable data.
A sufficient statistic T(X) is an observable random variable such that the conditional probability distribution of all observable data X given T(X) does not depend on any of the unobservable quantities
A Rao–Blackwell estimator d1(X) of an unobservable quantity ? is the conditional expected value E(d(X) | T(X)) of some estimator d(X) given a sufficient statistic T(X)
The theorem states that the mean squared error of the Rao–Blackwell estimator does not exceed that of the original estimator
4. Marginalizing the filter (Rao-Blackwellization) Particle filter target posterior:
Suppose we divide state into two groups and such that
and
is analytically tractable
Decompose posterior:
5. Generic Rao-Blackwellized Particle Filter
6. Partially analytical state update
This step seems not contain the term .However this is an intractable theoretical formulation. If we try to decompose, we encounter expensive integrals over
Mostly used method for this problem is approximating the conditional distribution over as a linear-Gaussian distribution.
7. Practical RBPF Algorithm Assumptions:
is independent of , conditioned on
We can define
are calculated from
Particles are , is covariance matrix of
8. Practical RBPF Algorithm
9. Mathematical Derivation
10. Mathematical Derivation